Pedestian Recognition in Low Illumination with Deep Neural Network and SVM
Computer vision development has been deployed in many areas e.g., self-driving cars, intelligent video tracking systems, military, medicine, and traffic surveillance. Successful object tracking is complex and depends on various factors e.g., environment and light sources. Real-time detection of objects at night or in difficult visibility conditions is still challenging despite the developed detection methods. Even though object detection and tracking methods have remarkably improved over the past few years and they provide very accurate detections, they are mainly designed for the daylight conditions. Object detection at night-time is complicated due to the low contrast between instances and the background and in the lack of colour differentiation it is complex to distinguish object shapes. In this research common detection methods were applied on frames captured in low visibility. To improve object detection the new method for detecting pedestrians in low illumination conditions using the deeplabv3 and SVM was introduced to verify the presence of pedestrians on sequenced frames. This approach is computationally economical and superior in comparison to common methods. Due to the efficiency and accuracy of this method it can be used for surveillance and security systems.
Keywords - Real Time Tracking, Object Recognition, Low Illumination